Alignment between Brains and AI: Evidence for Convergent Evolution across Modalities, Scales and Training Trajectories arxiv.org/abs/2507.01966

Alignment between Brains and AI: Evidence for Convergent Evolution across Modalities, Scales and Training Trajectories

Artificial and biological systems may evolve similar computational solutions despite fundamental differences in architecture and learning mechanisms -- a form of convergent evolution. We demonstrate this phenomenon through large-scale analysis of alignment between human brain activity and internal representations of over 600 AI models spanning language and vision domains, from 1.33M to 72B parameters. Analyzing 60 million alignment measurements reveals that higher-performing models spontaneously develop stronger brain alignment without explicit neural constraints, with language models showing markedly stronger correlation (r=0.89, p<7.5e-13) than vision models (r=0.53, p<2.0e-44). Crucially, longitudinal analysis demonstrates that brain alignment consistently precedes performance improvements during training, suggesting that developing brain-like representations may be a necessary stepping stone toward higher capabilities. We find systematic patterns: language models exhibit strongest alignment with limbic and integrative regions, while vision models show progressive alignment with visual cortices; deeper processing layers converge across modalities; and as representational scale increases, alignment systematically shifts from primary sensory to higher-order associative regions. These findings provide compelling evidence that optimization for task performance naturally drives AI systems toward brain-like computational strategies, offering both fundamental insights into principles of intelligent information processing and practical guidance for developing more capable AI systems.

arXiv.org

Ghost in the Machine: Examining the Philosophical Implications of Recursive Algorithms in Artificial Intelligence Systems arxiv.org/abs/2507.01967

Ghost in the Machine: Examining the Philosophical Implications of Recursive Algorithms in Artificial Intelligence Systems

This paper investigates whether contemporary AI architectures employing deep recursion, meta-learning, and self-referential mechanisms provide evidence of machine consciousness. Integrating philosophical history, cognitive science, and AI engineering, it situates recursive algorithms within a lineage spanning Cartesian dualism, Husserlian intentionality, Integrated Information Theory, the Global Workspace model, and enactivist perspectives. The argument proceeds through textual analysis, comparative architecture review, and synthesis of neuroscience findings on integration and prediction. Methodologically, the study combines conceptual analysis, case studies, and normative risk assessment informed by phenomenology and embodied cognition. Technical examples, including transformer self-attention, meta-cognitive agents, and neuromorphic chips, illustrate how functional self-modeling can arise without subjective experience. By distinguishing functional from phenomenal consciousness, the paper argues that symbol grounding, embodiment, and affective qualia remain unresolved barriers to attributing sentience to current AI. Ethical analysis explores risks of premature anthropomorphism versus neglect of future sentient systems; legal implications include personhood, liability, authorship, and labor impacts. Future directions include quantum architectures, embodied robotics, unsupervised world modeling, and empirical tests for non-biological phenomenality. The study reframes the "hard problem" as a graded and increasingly testable phenomenon, rather than a metaphysical impasse. It concludes that recursive self-referential design enhances capability but does not entail consciousness or justify moral status. Keywords: Recursive algorithms; self-reference; machine consciousness; AI ethics; AI consciousness

arXiv.org

TubuleTracker: a high-fidelity shareware software to quantify angiogenesis architecture and maturity arxiv.org/abs/2507.02024

TubuleTracker: a high-fidelity shareware software to quantify angiogenesis architecture and maturity

Background: In vitro endothelial cell culture is widely used to study angiogenesis. Histomicrographic images of cell networks are often analyzed manually, a process that is time-consuming and subjective. Automated tools like ImageJ (NIH) can assist, but are often slow and inaccurate. Additionally, as endothelial networks grow more complex, traditional architectural metrics may not fully reflect network maturity. To address these limitations, we developed tubuleTracker, a software tool that quantifies endothelial network architecture and maturity rapidly and objectively. Methods: Human umbilical vein endothelial cells were cultured in an extracellular matrix, and 54 images were acquired using phase contrast microscopy. Each image was analyzed manually by three independent reviewers, and by both ImageJ and tubuleTracker. Key metrics included tubule count, total length, node count, tubule area, and vessel circularity. In parallel, trained scientists rated each image for angiogenesis maturity on a 1-5 scale (1 = most mature). Results: Analysis time per image differed significantly: manual (8 min), ImageJ (58+/-4 s), and tubuleTracker (6+/-2 s) (p<0.0001). Significant differences were also found in tubule count (manual 168+/-SD, tubuleTracker 92+/-SD, ImageJ 433+/-SD), length, and node count (all p<0.0001). tubuleTracker's metrics varied significantly across angiogenesis maturity scores, including tubule count, length, node count, area, and circularity (all p<0.0001). Conclusions: tubuleTracker was faster and more consistent than both manual and ImageJ-based analysis. Vessel circularity proved especially effective in capturing angiogenesis maturity. tubuleTracker is available as free shareware for the biomedical research community.

arXiv.org

REMI: Reconstructing Episodic Memory During Intrinsic Path Planning arxiv.org/abs/2507.02064

REMI: Reconstructing Episodic Memory During Intrinsic Path Planning

Grid cells in the medial entorhinal cortex (MEC) are believed to path integrate speed and direction signals to activate at triangular grids of locations in an environment, thus implementing a population code for position. In parallel, place cells in the hippocampus (HC) fire at spatially confined locations, with selectivity tuned not only to allocentric position but also to environmental contexts, such as sensory cues. Although grid and place cells both encode spatial information and support memory for multiple locations, why animals maintain two such representations remains unclear. Noting that place representations seem to have other functional roles in intrinsically motivated tasks such as recalling locations from sensory cues, we propose that animals maintain grid and place representations together to support planning. Specifically, we posit that place cells auto-associate not only sensory information relayed from the MEC but also grid cell patterns, enabling recall of goal location grid patterns from sensory and motivational cues, permitting subsequent planning with only grid representations. We extend a previous theoretical framework for grid-cell-based planning and show that local transition rules can generalize to long-distance path forecasting. We further show that a planning network can sequentially update grid cell states toward the goal. During this process, intermediate grid activity can trigger place cell pattern completion, reconstructing experiences along the planned path. We demonstrate all these effects using a single-layer RNN that simultaneously models the HC-MEC loop and the planning subnetwork. We show that such recurrent mechanisms for grid cell-based planning, with goal recall driven by the place system, make several characteristic, testable predictions.

arXiv.org

Coexistence and Extinction in Flow-Kick Systems: An invasion growth rate approach arxiv.org/abs/2507.02157

Coexistence and Extinction in Flow-Kick Systems: An invasion growth rate approach

Populations experience a complex interplay of continuous and discrete processes: continuous growth and interactions are punctuated by discrete reproduction events, dispersal, and external disturbances. These dynamics can be modeled by impulsive or flow-kick systems, where continuous flows alternate with instantaneous discrete changes. To study species persistence in these systems, an invasion growth rate theory is developed for flow-kick models with state-dependent timing of kicks and auxiliary variables. The invasion growth rates are Lyapunov exponents characterizing the average per-capita growth of species when rare. Two theorems are proven to characterize permanence i.e. the extinction set is a repellor. The first theorem uses Morse decompositions of the extinction set and requires that there exists a species with a positive invasion growth rate for every invariant measure supported on a component of the Morse decomposition. The second theorem uses invasion growth rates to define invasion graphs whose vertices correspond to communities and directed edges to potential invasions. Provided the invasion graph is acyclic, permanence is fully characterized by the signs of the invasion growth rates. Invasion growth rates are also used to identify the existence of extinction-bound trajectories and attractors that lie on the extinction set. The results are illustrated with three applications: (i) a microbial serial transfer model, (ii) a spatially structured consumer-resource model, and (iii) an empirically parameterized Lotka-Volterra model. Mathematical challenges and promising biological applications are discussed.

arXiv.org

A Multi-Scale Finite Element Method for Investigating Fiber Remodeling in Hypertrophic Cardiomyopathy arxiv.org/abs/2507.02193

A Multi-Scale Finite Element Method for Investigating Fiber Remodeling in Hypertrophic Cardiomyopathy

A significant hallmark of hypertrophic cardiomyopathy (HCM) is fiber disarray, which is associated with various cardiac events such as heart failure. Quantifying fiber disarray remains critical for understanding the disease s complex pathophysiology. This study investigates the role of heterogeneous HCM-induced cellular abnormalities in the development of fiber disarray and their subsequent impact on cardiac pumping function. Fiber disarray is predicted using a stress-based law to reorient myofibers and collagen within a multiscale finite element cardiac modeling framework, MyoFE. Specifically, the model is used to quantify the distinct impacts of heterogeneous distributions of hypercontractility, hypocontractility, and fibrosis on fiber disarray development and examines their effect on functional characteristics of the heart. Our results show that heterogenous cell level abnormalities highly disrupt the normal mechanics of myocardium and lead to significant fiber disarray. The pattern of disarray varies depending on the specific perturbation, offering valuable insights into the progression of HCM. Despite the random distribution of perturbed regions within the cardiac muscle, significantly higher fiber disarray is observed near the epicardium compared to the endocardium across all perturbed left ventricle (LV) models. This regional difference in fiber disarray, irrespective of perturbation severity, aligns with previous DT-MRI studies, highlighting the role of regional myocardial mechanics in the development of fiber disarray. Furthermore, cardiac performance declined in the remodeled LVs, particularly in those with fibrosis and hypocontractility. These findings provide important insights into the structural and functional consequences of HCM and offer a framework for future investigations into therapeutic interventions targeting cardiac remodeling.

arXiv.org

NLP4Neuro: Sequence-to-sequence learning for neural population decoding arxiv.org/abs/2507.02264

NLP4Neuro: Sequence-to-sequence learning for neural population decoding

Delineating how animal behavior arises from neural activity is a foundational goal of neuroscience. However, as the computations underlying behavior unfold in networks of thousands of individual neurons across the entire brain, this presents challenges for investigating neural roles and computational mechanisms in large, densely wired mammalian brains during behavior. Transformers, the backbones of modern large language models (LLMs), have become powerful tools for neural decoding from smaller neural populations. These modern LLMs have benefited from extensive pre-training, and their sequence-to-sequence learning has been shown to generalize to novel tasks and data modalities, which may also confer advantages for neural decoding from larger, brain-wide activity recordings. Here, we present a systematic evaluation of off-the-shelf LLMs to decode behavior from brain-wide populations, termed NLP4Neuro, which we used to test LLMs on simultaneous calcium imaging and behavior recordings in larval zebrafish exposed to visual motion stimuli. Through NLP4Neuro, we found that LLMs become better at neural decoding when they use pre-trained weights learned from textual natural language data. Moreover, we found that a recent mixture-of-experts LLM, DeepSeek Coder-7b, significantly improved behavioral decoding accuracy, predicted tail movements over long timescales, and provided anatomically consistent highly interpretable readouts of neuron salience. NLP4Neuro demonstrates that LLMs are highly capable of informing brain-wide neural circuit dissection.

arXiv.org

Finding Similar Objects and Active Inference for Surprise in Numenta Neocortex Model arxiv.org/abs/2506.21554

Finding Similar Objects and Active Inference for Surprise in Numenta Neocortex Model

Jeff Hawkins and his colleagues in Numenta have proposed the thousand-brains system. This is a model of the structure and operation of the neocortex and is under investigation as a new form of artificial intelligence. In their study, learning and inference algorithms running on the system are proposed, where the prediction is an important function. The author believes that one of the most important capabilities of the neocortex in addition to prediction is the ability to make association, that is, to find the relationships between objects. Similarity is an important example of such relationships. In our study, algorithms that run on the thousand-brains system to find similarities are proposed. Although the setting for these algorithms is restricted, the author believes that the case it covers is fundamental. Karl Friston and his colleagues have studied the free-energy principle that explains how the brain actively infers the cause of a Shannon surprise. In our study, an algorithm is proposed for the thousand-brains system to make this inference. The problem of inferring what is being observed from the sensory data is a type of inverse problem, and the inference algorithms of the thousand-brains system and free-energy principle solve this problem in a Bayesian manner. Our inference algorithms can also be interpreted as Bayesian or non-Bayesian updating processes.

arXiv.org

Quantum Variational Transformer Model for Enhanced Cancer Classification arxiv.org/abs/2506.21641

Quantum Variational Transformer Model for Enhanced Cancer Classification

Accurate prediction of cancer type and primary tumor site is critical for effective diagnosis, personalized treatment, and improved outcomes. Traditional models struggle with the complexity of genomic and clinical data, but quantum computing offers enhanced computational capabilities. This study develops a hybrid quantum-classical transformer model, incorporating quantum attention mechanisms via variational quantum circuits (VQCs) to improve prediction accuracy. Using 30,000 anonymized cancer samples from the Genome Warehouse (GWH), data preprocessing included cleaning, encoding, and feature selection. Classical self-attention modules were replaced with quantum attention layers, with classical data encoded into quantum states via amplitude encoding. The model, trained using hybrid backpropagation and quantum gradient calculations, outperformed the classical transformer model, achieving 92.8% accuracy and an AUC of 0.96 compared to 87.5% accuracy and an AUC of 0.89. It also demonstrated 35% faster training and 25% fewer parameters, highlighting computational efficiency. These findings showcase the potential of quantum-enhanced transformers to advance biomedical data analysis, enabling more accurate diagnostics and personalized medicine.

arXiv.org

The Spatiotemporal Organization of Motor Cortex Activity Supporting Manual Dexterity arxiv.org/abs/2506.21738

The Spatiotemporal Organization of Motor Cortex Activity Supporting Manual Dexterity

Motor cortex (M1) is a crucial brain area for controlling voluntary movements, such as reaching and grasping for a cup of coffee. M1 is organized in a somatotopic manner, such that M1 output driving movement to different parts of the body is organized along the cortical surface. In primates, the arm and hand are represented in M1 as separate but overlapping territories. Unit activity recorded from the M1 forelimb representation comodulates with parameters related to reaching and/or grasping. The overall aim of this dissertation is to understand the spatiotemporal dynamics of M1 activity that produces reach-to-grasp movements. To address this goal, intracortical microstimulation (ICMS) is delivered along the precentral gyrus of two macaque monkeys to define the M1 motor map. Subsequently, cortical activity is recorded from the M1 forelimb representation using intrinsic signal optical imaging (ISOI) while macaques execute an instructed reach-to-grasp task. Results from imaging experiments produce spatial maps that define cortical territories with increased activity during reach-to-grasp movements. Next, unit activity was recorded from the M1 forelimb representation with a laminar multielectrode while macaques completed the same reach-to-grasp task. Recording site locations differed between sessions to comprehensively sample unit responses throughout the M1 forelimb representation. Imaging experiments reveal that activity supporting reach-to-grasp movements was concentrated in patches that comprise less than half of the M1 forelimb representation. Electrophysiology recordings reveal that activity related to reaching is spatially organized within M1 distinctly from activity related to grasping. The results support the idea that spatial organizing principles are inherent in M1 activity that supports reach-to-grasp movements.

arXiv.org

Vegetation Patterning Can Both Impede and Trigger Critical Transitions from Savanna to Grassland arxiv.org/abs/2506.22178

Vegetation Patterning Can Both Impede and Trigger Critical Transitions from Savanna to Grassland

Tree-grass coexistence is a defining feature of savanna ecosystems, which play an important role in supporting biodiversity and human populations worldwide. While recent advances have clarified many of the underlying processes, how these mechanisms interact to shape ecosystem dynamics under environmental stress is not yet understood. Here, we present and analyze a minimalistic spatially extended model of tree-grass dynamics in dry savannas. We incorporate tree facilitation of grasses through shading and grass competing with trees for water, both varying with tree life stage. Our model shows that these mechanisms lead to grass-tree coexistence and bistability between savanna and grassland states. Moreover, the model predicts vegetation patterns consisting of trees and grasses, particularly under harsh environmental conditions, which can persist in situations where a non-spatial version of the model predicts ecosystem collapse from savanna to grassland instead (a phenomenon called ''Turing-evades-tipping''). Additionally, we identify a novel ''Turing-triggers-tipping'' mechanism, where unstable pattern formation drives tipping events that are overlooked when spatial dynamics are not included. These transient patterns act as early warning signals for ecosystem transitions, offering a critical window for intervention. Further theoretical and empirical research is needed to determine when spatial patterns prevent tipping or drive collapse.

arXiv.org

Droplet growth, Ostwald's rule, and emergence of order in Fused in Sarcoma arxiv.org/abs/2506.21792 -mat.soft

Droplet growth, Ostwald's rule, and emergence of order in Fused in Sarcoma

The low complexity domain of Fused in Sarcoma (FUS-LC consisting of 214 residues) undergoes phase separation, resulting in a dense liquid-like phase that forms early and slowly matures to reach ordered gel-like state on long time scales. Upon maturation, core-1, comprising of the 57 residues (39-95) in the N-terminus become structured, resulting in the formation of a non-polymorphic fibril. The truncated FUS-LC-C (residues 110-214) construct forms a fibril in which core-2 (residues 112-150) adopts a $β$-sheet structure. Using coarse-grained monomer SOP-IDP model simulations of FUS-LC, we predict that residues 155-190 in the C-terminal (core-3) form rapidly, followed by core-2, and finally core-1. The time scale of formation of the cores and their stabilities are inversely correlated, as anticipated by the Ostwald's rule of stages. Unbiased multichain simulations show that the chemical potentials in the two phases are equal and the calculated densities of the dense and dilute phases are in agreement with experiments. The dense phase, which forms by a nucleation mechanism, coarsens over time by a process that is reminiscent of Ostwald ripening. AlphaFold predictions of the core-3 structure and the simulations show that $β$-strand emerges in the core-3 region early during the droplet formation, and drives the initiation of FUS-LC assembly. The techniques introduced here are general and could be used to probe assembly of other IDPs such as TDP-43, which shares many features with FUS-LC.

arXiv.org

Single-Trajectory Bayesian Modeling Reveals Multi-State Diffusion of the MSH Sliding Clamp arxiv.org/abs/2506.21943

Single-Trajectory Bayesian Modeling Reveals Multi-State Diffusion of the MSH Sliding Clamp

DNA mismatch repair (MMR) is the essential mechanism for preserving genomic integrity in various living organisms. In this process, MutS homologs (MSH) play crucial roles in identifying mismatched basepairs and recruiting downstream MMR proteins. The MSH protein exhibits distinct functions and diffusion dynamics before and after the recognition of mismatches while traversing along DNA. An ADP-bound MSH, known as the MSH searching clamp, scans DNA sequences via rotational diffusion along the DNA backbone. Upon recognizing a mismatch, the MSH combines with ATP molecules, forming a stable sliding clamp. Recent experimental evidence challenges the conventional view that the sliding clamp performs a simple Brownian motion. In this study, we explore the diffusion dynamics of the ATP-bound MSH sliding clamp through single-particle tracking experiments and introduce a Bayesian single-trajectory modeling framework to analyze its motion. Our quantitative analysis reveals that the diffusion characteristics defy explanation by a single-state diffusion mechanism. Instead, our in-depth model inference uncovers three distinct diffusion states, each characterized by specific diffusion coefficients. These states alternate over time, with cross-state transitions predominantly involving one intermediate state, and direct transitions between the slowest and the fastest states being scarce. We propose that these multi-state dynamics reflect underlying conformational changes in the MSH sliding clamp, highlighting a more intricate diffusion mechanism than previously appreciated.

arXiv.org
Show older
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.